Catenary

Catenary is a desktop IDE built around an infinite, zoomable canvas where terminals, editors, browsers, and AI coding agents float side by side. Created by Giorgio Nícolas de Arruda Bertino, it is designed specifically for running multiple AI coding agents simultaneously — Claude Code, Codex, Cursor, Grok, OpenCode, and Pi — without losing track of what each one is doing. The app is local-first by construction: no account, no telemetry, and no code ever leaves your machine.

What It Is

Catenary is a native Electron + Chromium desktop application that opens a local folder and provides an infinite canvas for spatial arrangement of development panels. It is not a cloud service and not an AI agent itself — it orchestrates the CLI-based agents you already have installed, using the API keys you already pay for. Canvas state is stored locally in .cate/workspace.json and .cate/session.json, so everything comes back exactly where you left it when you reopen a project.

Core Canvas Features

The canvas is the central metaphor: every panel — terminal, Monaco editor, browser, git worktree — floats in canvas space and can be panned, zoomed, and dragged freely. Key capabilities include:

• 1-Click worktrees & branches: The "New Agent Task" button creates an isolated branch with its own working directory and agent, color-coded throughout the sidebar and canvas, without touching the main working tree.
• Monaco with diffs: VS Code's editor is embedded in the canvas with side-by-side diffs of everything an agent touched, alongside a git-aware file tree.
• Context wires: Directed visual lines connect panels to agents, relaying context automatically when a source goes quiet or acting as standing permissions for downstream agents to pull from.
• Maestro Mode: One agent recruits, briefs, and wires a team of helpers (up to five by default, adjustable to ten). Every action Maestro takes opens an editable approval card before anything appears on the canvas.
• Multi-project & parallel branches: Each open project keeps its own canvas; tasks run in isolated worktrees with their own agents, and switching between projects restores state completely.

Privacy Architecture

Catenary's privacy model is unusually explicit. The homepage lists every outbound network call the app makes — exactly two hosts:

• A GET to github.com/GioNicolas/catenary-releases every 15 minutes to check for updates (no data sent, only a version number received).
• A POST to thecatenary.app/api/license/activate, /deactivate, or /api/trial/claim — only triggered by user action when pasting a key, removing it, or claiming the trial. The only data sent is a SHA-256 machine ID (64 hex chars, irreversible, no name or email).

Source code, terminal output, file paths, prompts, agent conversations, and API keys are never sent anywhere. The analytics endpoint ships empty in every build, and the crash-reporter DSN was removed from the distribution entirely. The Pro license is verified offline via Ed25519 signature checked locally, so the app remains fully functional without an internet connection.

Supported Agents & Platform

Catenary supports Windows, macOS (Apple Silicon and Intel), and Linux (AppImage, .deb, .tar.gz). It works with any agent CLI the user has installed — Claude Code, Codex, Cursor, Grok, OpenCode, Pi, or local models via Ollama or LM Studio. The app does not proxy prompts; agents talk directly to Anthropic, OpenAI, Google, or a local endpoint using the user's own API keys.

Update: v1.0.4

The GitHub Releases repository (GioNicolas/catenary-releases) shows the latest published version as v1.0.4, released on 2026-09-07. The plan limits document was last updated on 2026-09-04 (with a prior version on 2026-08-28), indicating active early development. The installer is not yet code-signed, which the documentation acknowledges explicitly with instructions for bypassing the OS security prompt on first run. The 1.x line is the current release track, and Pro license holders receive updates within this line.